PDC Sensor for Person Detection - Ultrasonic Doppler and Time-of-Flight Sensing for Human Presence and Activity Recognition
This in-depth technical article examines the application of PDC sensors for person detection, covering the ultrasonic Doppler and time-of-flight sensing principles, the multi-stage classification algorithms for activity recognition, the sensor network configuration for indoor monitoring, and the integration with building automation and safety systems.
Ultrasonic person detection employs two primary sensing modalities: Doppler shift detection for moving persons and time-of-flight measurement for presence and positioning. The Doppler method exploits the frequency shift of reflected ultrasonic signals caused by moving objects, such as a walking person, where the reflected signal frequency differs from the transmitted frequency. This Doppler shift only arises when there is movement close by, providing a direct indication of human presence. The time-of-flight method measures the distance to static or moving objects by emitting ultrasonic pulses and measuring the echo return time. For person detection, the received signal waveforms differ from one frame to the next when a moving person is present, requiring the recording of changes at intervals over time with significant arithmetic processing. Multi-stage classification processes have been developed to discriminate between different human activities in indoor environments using ultrasonic sensor networks. The raw ultrasonic sensor data undergoes pre-processing, including calibration to filter out environmental noise and segmentation into time windows (typically 5 seconds), transforming raw distance data into meaningful displacement measurements.
PDC Sensor
The multi-stage classification approach for activity recognition combines machine learning and deep learning techniques to achieve high accuracy. In the first stage, a support vector machine (SVM) classifier distinguishes between static and dynamic office activities with 93.1% accuracy. In the second stage, a convolutional neural network further classifies specific dynamic office activities, such as writing, typing on a PC, talking on the phone, and standing, with a mean accuracy of 99.3%. The root mean square temporal marker is extracted and used as an input feature in this two-stage classification approach to differentiate between low-intensity and high-intensity activities. This scalable solution has the potential to enhance user comfort by integrating ultrasonic-based human activity recognition systems into personalized comfort models and optimizing resource usage in indoor environments. The non-invasive and non-wearable nature of the system ensures that it does not disturb occupants while providing reliable activity monitoring.
The sensor network configuration for indoor person detection typically involves multiple ultrasonic sensors distributed throughout the monitored space. The sensors can be ceiling-mounted or wall-mounted to provide comprehensive coverage of the area. The sensor network operates in a coordinated manner, with each sensor measuring distance to the nearest object and transmitting data to a central processing unit. The system uses the principle of multiple reflections of ultrasonic waves at the same obstacle edge to enable precise edge detection and positioning. The occupancy status of each cell in a grid representation of the monitored area is determined based on the distance measurements, enabling the identification of occupied spaces. The system can also employ supervised learning algorithms to estimate the number of occupants in a room based on the collected sensor data. The algorithm based on a particle filter allows for more information to be extracted from the relatively simple sensor signal by detecting human walking movement specifically, opening up the way for an ultrasound-based indoor positioning system.
The integration with building automation systems uses person detection data to optimize lighting, heating, ventilation, and air conditioning (HVAC) operations. By detecting the presence and number of occupants in a room, the building management system can adjust lighting levels and temperature settings to reduce energy consumption when spaces are unoccupied. The system can also be used for security applications, detecting unauthorized presence in restricted areas. In emergency situations, such as smoke-filled buildings, ultrasonic sensors can assist fire brigades by detecting human presence and directing them to locations where help is most needed. The sensors are preferred for these applications due to their wide detection area, less light dependency, ability to detect glass and shiny walls, small size, lightweight, low memory usage, and lower power consumption compared to laser range finders or cameras. The system can also be used for human following robots, leveraging the ultrasonic sensor's ability to detect persons regardless of lighting conditions.
The future of ultrasonic person detection is moving toward enhanced accuracy and integration with artificial intelligence. The use of supervised learning over segmented ultrasonic reflection patterns has been shown capable of detecting minute variations in the environment's response, enabling the detection of still persons who might otherwise be missed by Doppler-based methods. The development of hardware-efficient ultrasonic entrance counting systems using machine learning approaches is enabling accurate people counting for facility management and security applications. The integration of ultrasonic sensors with other sensing modalities, such as passive infrared (PIR) and radar, is improving the robustness of person detection in challenging environments. The ongoing advancement in signal processing and machine learning is making ultrasonic person detection more reliable and versatile, expanding its applications in smart buildings, security, and human-robot interaction.
